| 1 | #!/usr/bin/env python2
|
| 2 | """
|
| 3 | lazylex/html.py - Low-Level HTML Processing.
|
| 4 |
|
| 5 | See lazylex/README.md for details.
|
| 6 |
|
| 7 | TODO: This should be an Oils library eventually. It's a "lazily-parsed data
|
| 8 | structure" like TSV8
|
| 9 | """
|
| 10 | from __future__ import print_function
|
| 11 |
|
| 12 | import cStringIO
|
| 13 | import re
|
| 14 | import sys
|
| 15 |
|
| 16 | from typing import List, Tuple
|
| 17 |
|
| 18 |
|
| 19 | def log(msg, *args):
|
| 20 | msg = msg % args
|
| 21 | print(msg, file=sys.stderr)
|
| 22 |
|
| 23 |
|
| 24 | class LexError(Exception):
|
| 25 | """For bad lexical elements like <> or &&"""
|
| 26 |
|
| 27 | def __init__(self, s, pos):
|
| 28 | self.s = s
|
| 29 | self.pos = pos
|
| 30 |
|
| 31 | def __str__(self):
|
| 32 | return '(LexError %r)' % (self.s[self.pos:self.pos + 20])
|
| 33 |
|
| 34 |
|
| 35 | class ParseError(Exception):
|
| 36 | """For errors in the tag structure."""
|
| 37 |
|
| 38 | def __init__(self, msg, *args):
|
| 39 | self.msg = msg
|
| 40 | self.args = args
|
| 41 |
|
| 42 | def __str__(self):
|
| 43 | return '(ParseError %s)' % (self.msg % self.args)
|
| 44 |
|
| 45 |
|
| 46 | class Output(object):
|
| 47 | """Takes an underlying input buffer and an output file. Maintains a
|
| 48 | position in the input buffer.
|
| 49 |
|
| 50 | Print FROM the input or print new text to the output.
|
| 51 | """
|
| 52 |
|
| 53 | def __init__(self, s, f, left_pos=0, right_pos=-1):
|
| 54 | self.s = s
|
| 55 | self.f = f
|
| 56 | self.pos = left_pos
|
| 57 | self.right_pos = len(s) if right_pos == -1 else right_pos
|
| 58 |
|
| 59 | def SkipTo(self, pos):
|
| 60 | """Skip to a position."""
|
| 61 | self.pos = pos
|
| 62 |
|
| 63 | def PrintUntil(self, pos):
|
| 64 | """Print until a position."""
|
| 65 | piece = self.s[self.pos:pos]
|
| 66 | self.f.write(piece)
|
| 67 | self.pos = pos
|
| 68 |
|
| 69 | def PrintTheRest(self):
|
| 70 | """Print until the end of the string."""
|
| 71 | self.PrintUntil(self.right_pos)
|
| 72 |
|
| 73 | def Print(self, s):
|
| 74 | """Print text to the underlying buffer."""
|
| 75 | self.f.write(s)
|
| 76 |
|
| 77 |
|
| 78 | # HTML Tokens
|
| 79 | TOKENS = 'Decl Comment Processing StartTag StartEndTag EndTag DecChar HexChar CharEntity RawData Invalid EndOfStream'.split(
|
| 80 | )
|
| 81 |
|
| 82 |
|
| 83 | class Tok(object):
|
| 84 | """
|
| 85 | Avoid lint errors by using these aliases
|
| 86 | """
|
| 87 | pass
|
| 88 |
|
| 89 |
|
| 90 | assert len(TOKENS) == 12, TOKENS
|
| 91 |
|
| 92 | TOKEN_NAMES = [None] * len(TOKENS) # type: List[str]
|
| 93 |
|
| 94 | this_module = sys.modules[__name__]
|
| 95 | for i, tok_str in enumerate(TOKENS):
|
| 96 | setattr(this_module, tok_str, i)
|
| 97 | setattr(Tok, tok_str, i)
|
| 98 | TOKEN_NAMES[i] = tok_str
|
| 99 |
|
| 100 |
|
| 101 | def TokenName(tok_id):
|
| 102 | return TOKEN_NAMES[tok_id]
|
| 103 |
|
| 104 |
|
| 105 | def MakeLexer(rules):
|
| 106 | return [
|
| 107 | # DOTALL is for the comment
|
| 108 | (re.compile(pat, re.VERBOSE | re.DOTALL), i) for (pat, i) in rules
|
| 109 | ]
|
| 110 |
|
| 111 |
|
| 112 | #
|
| 113 | # Eggex
|
| 114 | #
|
| 115 | # Tag = / ~['>']+ /
|
| 116 |
|
| 117 | # Is this valid? A single character?
|
| 118 | # Tag = / ~'>'* /
|
| 119 |
|
| 120 | # Maybe better: / [NOT '>']+/
|
| 121 | # capital letters not allowed there?
|
| 122 | #
|
| 123 | # But then this is confusing:
|
| 124 | # / [NOT ~digit]+/
|
| 125 | #
|
| 126 | # / [NOT digit] / is [^\d]
|
| 127 | # / ~digit / is \D
|
| 128 | #
|
| 129 | # Or maybe:
|
| 130 | #
|
| 131 | # / [~ digit]+ /
|
| 132 | # / [~ '>']+ /
|
| 133 | # / [NOT '>']+ /
|
| 134 |
|
| 135 | # End = / '</' Tag '>' /
|
| 136 | # StartEnd = / '<' Tag '/>' /
|
| 137 | # Start = / '<' Tag '>' /
|
| 138 | #
|
| 139 | # EntityRef = / '&' dot{* N} ';' /
|
| 140 |
|
| 141 | LEXER = [
|
| 142 | # TODO: instead of nongreedy matches, the loop can just do .find('-->') and
|
| 143 | # .find('?>')
|
| 144 |
|
| 145 | # Actually non-greedy matches are regular and can be matched in linear time
|
| 146 | # with RE2.
|
| 147 | #
|
| 148 | # https://news.ycombinator.com/item?id=27099798
|
| 149 | #
|
| 150 | # Maybe try combining all of these for speed.
|
| 151 | (r'<!-- .*? -->', Tok.Comment),
|
| 152 | (r'<\? .*? \?>', Tok.Processing),
|
| 153 |
|
| 154 | # NOTE: < is allowed in these.
|
| 155 | (r'<! [^>]+ >', Tok.Decl), # <!DOCTYPE html>
|
| 156 | (r'</ [^>]+ >', Tok.EndTag), # self-closing <br/> comes FIRST
|
| 157 | (r'< [^>]+ />', Tok.StartEndTag), # end </a>
|
| 158 | (r'< [^>]+ >', Tok.StartTag), # start <a>
|
| 159 | (r'&\# [0-9]+ ;', Tok.DecChar),
|
| 160 | (r'&\# x[0-9a-fA-F]+ ;', Tok.HexChar),
|
| 161 | (r'& [a-zA-Z]+ ;', Tok.CharEntity),
|
| 162 |
|
| 163 | # Note: > is allowed in raw data.
|
| 164 | # https://stackoverflow.com/questions/10462348/right-angle-bracket-in-html
|
| 165 | (r'[^&<]+', Tok.RawData),
|
| 166 | (r'.', Tok.Invalid), # error!
|
| 167 | ]
|
| 168 |
|
| 169 | LEXER = MakeLexer(LEXER)
|
| 170 |
|
| 171 |
|
| 172 | class Lexer(object):
|
| 173 |
|
| 174 | def __init__(self, s, left_pos=0, right_pos=-1):
|
| 175 | self.s = s
|
| 176 | self.pos = left_pos
|
| 177 | self.right_pos = len(s) if right_pos == -1 else right_pos
|
| 178 | self.cache = {} # string -> compiled regex pattern object
|
| 179 |
|
| 180 | def _Peek(self):
|
| 181 | # type: () -> Tuple[int, int]
|
| 182 | """
|
| 183 | Note: not using _Peek() now
|
| 184 | """
|
| 185 | if self.pos == self.right_pos:
|
| 186 | return Tok.EndOfStream, self.pos
|
| 187 |
|
| 188 | assert self.pos < self.right_pos, self.pos
|
| 189 |
|
| 190 | # Find the first match.
|
| 191 | # Note: frontend/match.py uses _LongestMatch(), which is different!
|
| 192 | # TODO: reconcile them. This lexer should be expressible in re2c.
|
| 193 |
|
| 194 | # TODO: Get rid of non-greedy match
|
| 195 |
|
| 196 | for pat, tok_id in LEXER:
|
| 197 | m = pat.match(self.s, self.pos)
|
| 198 | if m:
|
| 199 | return tok_id, m.end()
|
| 200 | else:
|
| 201 | raise AssertionError('Tok.Invalid rule should have matched')
|
| 202 |
|
| 203 | def Read(self):
|
| 204 | # type: () -> Tuple[int, int]
|
| 205 | tok_id, end_pos = self._Peek()
|
| 206 | self.pos = end_pos # advance
|
| 207 | return tok_id, end_pos
|
| 208 |
|
| 209 | def LookAhead(self, regex):
|
| 210 | # Cache the regex compilation. This could also be LookAheadFor(THEAD)
|
| 211 | # or something.
|
| 212 | pat = self.cache.get(regex)
|
| 213 | if pat is None:
|
| 214 | pat = re.compile(regex)
|
| 215 | self.cache[regex] = pat
|
| 216 |
|
| 217 | m = pat.match(self.s, self.pos)
|
| 218 | return m is not None
|
| 219 |
|
| 220 |
|
| 221 | def _Tokens(s, left_pos, right_pos):
|
| 222 | """
|
| 223 | Args:
|
| 224 | s: string to parse
|
| 225 | left_pos, right_pos: Optional span boundaries.
|
| 226 | """
|
| 227 | lx = Lexer(s, left_pos, right_pos)
|
| 228 | while True:
|
| 229 | tok_id, pos = lx.Read()
|
| 230 | yield tok_id, pos
|
| 231 | if tok_id == Tok.EndOfStream:
|
| 232 | break
|
| 233 |
|
| 234 |
|
| 235 | def ValidTokens(s, left_pos=0, right_pos=-1):
|
| 236 | """Wrapper around _Tokens to prevent callers from having to handle Invalid.
|
| 237 |
|
| 238 | I'm not combining the two functions because I might want to do a
|
| 239 | 'yield' transformation on Tokens()? Exceptions might complicate the
|
| 240 | issue?
|
| 241 | """
|
| 242 | pos = left_pos
|
| 243 | for tok_id, end_pos in _Tokens(s, left_pos, right_pos):
|
| 244 | if tok_id == Tok.Invalid:
|
| 245 | raise LexError(s, pos)
|
| 246 | yield tok_id, end_pos
|
| 247 | pos = end_pos
|
| 248 |
|
| 249 |
|
| 250 | # Tag names:
|
| 251 | # Match <a or </a
|
| 252 | # Match <h2, but not <2h
|
| 253 | #
|
| 254 | # HTML 5 doesn't restrict tag names at all
|
| 255 | # https://html.spec.whatwg.org/#toc-syntax
|
| 256 | #
|
| 257 | # XML allows : - .
|
| 258 | # https://www.w3.org/TR/xml/#NT-NameChar
|
| 259 |
|
| 260 | # Namespaces for MathML, SVG
|
| 261 | # XLink, XML, XMLNS
|
| 262 | #
|
| 263 | # https://infra.spec.whatwg.org/#namespaces
|
| 264 | #
|
| 265 | # Allow - for td-attrs
|
| 266 |
|
| 267 | _TAG_RE = re.compile(r'/? \s* ([a-zA-Z][a-zA-Z0-9-]*)', re.VERBOSE)
|
| 268 |
|
| 269 | # To match href="foo"
|
| 270 |
|
| 271 | _ATTR_RE = re.compile(
|
| 272 | r'''
|
| 273 | \s+ # Leading whitespace is required
|
| 274 | ([a-z]+) # Attribute name
|
| 275 | (?: # Optional attribute value
|
| 276 | \s* = \s*
|
| 277 | (?:
|
| 278 | " ([^>"]*) " # double quoted value
|
| 279 | | ([a-zA-Z0-9_\-]+) # Just allow unquoted "identifiers"
|
| 280 | # TODO: relax this? for href=$foo
|
| 281 | )
|
| 282 | )?
|
| 283 | ''', re.VERBOSE)
|
| 284 |
|
| 285 | TagName, AttrName, UnquotedValue, QuotedValue = range(4)
|
| 286 |
|
| 287 |
|
| 288 | class TagLexer(object):
|
| 289 | """
|
| 290 | Given a tag like <a href="..."> or <link type="..." />, the TagLexer
|
| 291 | provides a few operations:
|
| 292 |
|
| 293 | - What is the tag?
|
| 294 | - Iterate through the attributes, giving (name, value_start_pos, value_end_pos)
|
| 295 | """
|
| 296 |
|
| 297 | def __init__(self, s):
|
| 298 | self.s = s
|
| 299 | self.start_pos = -1 # Invalid
|
| 300 | self.end_pos = -1
|
| 301 |
|
| 302 | def Reset(self, start_pos, end_pos):
|
| 303 | """Reuse instances of this object."""
|
| 304 | self.start_pos = start_pos
|
| 305 | self.end_pos = end_pos
|
| 306 |
|
| 307 | def TagString(self):
|
| 308 | return self.s[self.start_pos:self.end_pos]
|
| 309 |
|
| 310 | def TagName(self):
|
| 311 | # First event
|
| 312 | tok_id, start, end = next(self.Tokens())
|
| 313 | return self.s[start:end]
|
| 314 |
|
| 315 | def GetSpanForAttrValue(self, attr_name):
|
| 316 | # Algorithm: search for QuotedValue or UnquotedValue after AttrName
|
| 317 | # TODO: Could also cache these
|
| 318 |
|
| 319 | events = self.Tokens()
|
| 320 | val = (-1, -1)
|
| 321 | try:
|
| 322 | while True:
|
| 323 | tok_id, start, end = next(events)
|
| 324 | if tok_id == AttrName:
|
| 325 | name = self.s[start:end]
|
| 326 | if name == attr_name:
|
| 327 | # The value should come next
|
| 328 | tok_id, start, end = next(events)
|
| 329 | if tok_id in (QuotedValue, UnquotedValue):
|
| 330 | # Note: quoted values may have &
|
| 331 | # We would need ANOTHER lexer to unescape them.
|
| 332 | # Right now help_gen.py and oils_doc.py
|
| 333 | val = start, end
|
| 334 | break
|
| 335 |
|
| 336 | except StopIteration:
|
| 337 | pass
|
| 338 | return val
|
| 339 |
|
| 340 | def GetAttrRaw(self, attr_name):
|
| 341 | """
|
| 342 | Return the value, which may be UNESCAPED.
|
| 343 | """
|
| 344 | # Algorithm: search for QuotedValue or UnquotedValue after AttrName
|
| 345 | # TODO: Could also cache these
|
| 346 | start, end = self.GetSpanForAttrValue(attr_name)
|
| 347 | if start == -1:
|
| 348 | return None
|
| 349 | return self.s[start:end]
|
| 350 |
|
| 351 | def AllAttrsRaw(self):
|
| 352 | """
|
| 353 | Get a list of pairs [('class', 'foo'), ('href', '?foo=1&bar=2')]
|
| 354 |
|
| 355 | The quoted values may be escaped. We would need another lexer to
|
| 356 | unescape them.
|
| 357 | """
|
| 358 | pairs = []
|
| 359 | events = self.Tokens()
|
| 360 | try:
|
| 361 | while True:
|
| 362 | tok_id, start, end = next(events)
|
| 363 | if tok_id == AttrName:
|
| 364 | name = self.s[start:end]
|
| 365 |
|
| 366 | # The value should come next
|
| 367 | tok_id, start, end = next(events)
|
| 368 | if tok_id in (QuotedValue, UnquotedValue):
|
| 369 | # Note: quoted values may have &
|
| 370 | # We would need ANOTHER lexer to unescape them, but we
|
| 371 | # don't need that for ul-table
|
| 372 |
|
| 373 | val = self.s[start:end]
|
| 374 | pairs.append((name, val))
|
| 375 | except StopIteration:
|
| 376 | pass
|
| 377 | return pairs
|
| 378 |
|
| 379 | def Tokens(self):
|
| 380 | """
|
| 381 | Yields a sequence of tokens: Tag (AttrName AttrValue?)*
|
| 382 |
|
| 383 | Where each Token is (Type, start_pos, end_pos)
|
| 384 |
|
| 385 | Note that start and end are NOT redundant! We skip over some unwanted
|
| 386 | characters.
|
| 387 | """
|
| 388 | m = _TAG_RE.match(self.s, self.start_pos + 1)
|
| 389 | if not m:
|
| 390 | raise RuntimeError("Couldn't find HTML tag in %r" %
|
| 391 | self.TagString())
|
| 392 | yield TagName, m.start(1), m.end(1)
|
| 393 |
|
| 394 | pos = m.end(0)
|
| 395 |
|
| 396 | while True:
|
| 397 | # don't search past the end
|
| 398 | m = _ATTR_RE.match(self.s, pos, self.end_pos)
|
| 399 | if not m:
|
| 400 | # A validating parser would check that > or /> is next -- there's no junk
|
| 401 | break
|
| 402 |
|
| 403 | yield AttrName, m.start(1), m.end(1)
|
| 404 |
|
| 405 | # Quoted is group 2, unquoted is group 3.
|
| 406 | if m.group(2) is not None:
|
| 407 | yield QuotedValue, m.start(2), m.end(2)
|
| 408 | elif m.group(3) is not None:
|
| 409 | yield UnquotedValue, m.start(3), m.end(3)
|
| 410 |
|
| 411 | # Skip past the "
|
| 412 | pos = m.end(0)
|
| 413 |
|
| 414 |
|
| 415 | def ReadUntilStartTag(it, tag_lexer, tag_name):
|
| 416 | """Find the next <foo>, returning its (start, end) positions
|
| 417 |
|
| 418 | Raise ParseError if it's not found.
|
| 419 |
|
| 420 | tag_lexer is RESET.
|
| 421 | """
|
| 422 | pos = 0
|
| 423 | while True:
|
| 424 | try:
|
| 425 | tok_id, end_pos = next(it)
|
| 426 | except StopIteration:
|
| 427 | break
|
| 428 | tag_lexer.Reset(pos, end_pos)
|
| 429 | if tok_id == Tok.StartTag and tag_lexer.TagName() == tag_name:
|
| 430 | return pos, end_pos
|
| 431 |
|
| 432 | pos = end_pos
|
| 433 |
|
| 434 | raise ParseError('No start tag %r', tag_name)
|
| 435 |
|
| 436 |
|
| 437 | def ReadUntilEndTag(it, tag_lexer, tag_name):
|
| 438 | """Find the next </foo>, returning its (start, end) position
|
| 439 |
|
| 440 | Raise ParseError if it's not found.
|
| 441 |
|
| 442 | tag_lexer is RESET.
|
| 443 | """
|
| 444 | pos = 0
|
| 445 | while True:
|
| 446 | try:
|
| 447 | tok_id, end_pos = next(it)
|
| 448 | except StopIteration:
|
| 449 | break
|
| 450 | tag_lexer.Reset(pos, end_pos)
|
| 451 | if tok_id == Tok.EndTag and tag_lexer.TagName() == tag_name:
|
| 452 | return pos, end_pos
|
| 453 |
|
| 454 | pos = end_pos
|
| 455 |
|
| 456 | raise ParseError('No end tag %r', tag_name)
|
| 457 |
|
| 458 |
|
| 459 | CHAR_ENTITY = {
|
| 460 | 'amp': '&',
|
| 461 | 'lt': '<',
|
| 462 | 'gt': '>',
|
| 463 | 'quot': '"',
|
| 464 | }
|
| 465 |
|
| 466 |
|
| 467 | def ToText(s, left_pos=0, right_pos=-1):
|
| 468 | """Given HTML, return text by unquoting > and < etc.
|
| 469 |
|
| 470 | Used by:
|
| 471 | doctools/oils_doc.py: PygmentsPlugin
|
| 472 | doctool/make_help.py: HelpIndexCards
|
| 473 |
|
| 474 | In the latter case, we cold process some tags, like:
|
| 475 |
|
| 476 | - Blue Link (not clickable, but still useful)
|
| 477 | - Red X
|
| 478 |
|
| 479 | That should be html.ToAnsi.
|
| 480 | """
|
| 481 | f = cStringIO.StringIO()
|
| 482 | out = Output(s, f, left_pos, right_pos)
|
| 483 |
|
| 484 | pos = left_pos
|
| 485 | for tok_id, end_pos in ValidTokens(s, left_pos, right_pos):
|
| 486 | if tok_id == Tok.RawData:
|
| 487 | out.SkipTo(pos)
|
| 488 | out.PrintUntil(end_pos)
|
| 489 |
|
| 490 | elif tok_id == Tok.CharEntity: # &
|
| 491 |
|
| 492 | entity = s[pos + 1:end_pos - 1]
|
| 493 |
|
| 494 | out.SkipTo(pos)
|
| 495 | out.Print(CHAR_ENTITY[entity])
|
| 496 | out.SkipTo(end_pos)
|
| 497 |
|
| 498 | # Not handling these yet
|
| 499 | elif tok_id == Tok.HexChar:
|
| 500 | raise AssertionError('Hex Char %r' % s[pos:pos + 20])
|
| 501 |
|
| 502 | elif tok_id == Tok.DecChar:
|
| 503 | raise AssertionError('Dec Char %r' % s[pos:pos + 20])
|
| 504 |
|
| 505 | pos = end_pos
|
| 506 |
|
| 507 | out.PrintTheRest()
|
| 508 | return f.getvalue()
|